A large model driven intelligent scene automation design and construction system and method

Abstract

The application provides a large model driven intelligent scene automatic design and construction system and method, comprising a vehicle end, a cloud service platform and an execution module; the vehicle end collects environment perception data, cabin perception data, vehicle perception data, network perception data and user historical behavior data through a data acquisition module, and uploads the data to the cloud service platform through a data uploading module; the cloud service platform comprises a big data analysis layer and a large model layer, the big data analysis layer is used for data processing, feature extraction, user portrait construction and scene task triggering, the large model layer is used for language and intention understanding, scene reasoning planning, scene management and scene optimization, and a scene execution script is generated; the execution module calls vehicle atomization service to execute corresponding actions according to the scene execution script; feedback data after execution is used for updating the user portrait and optimizing the large model. The application can realize automatic construction, deployment, execution and continuous optimization of intelligent scenes.

Description

Technical Field

[0001] This invention relates to the fields of automotive intelligent cockpit, vehicle networking and artificial intelligence technology, and in particular to a large model-driven intelligent scene automated design and construction system and method. Background Technology

[0002] As vehicles become increasingly intelligent, the vehicle cockpit system is evolving from a traditional pre-defined function execution terminal into an intelligent cockpit system capable of sensing the environment, understanding users, learning behavior, and proactively providing services. Functions in an intelligent cockpit, such as air conditioning, seats, windows, lights, media playback, navigation, and driving modes, can typically be broken down into multiple individually callable atomic services. These services can then be combined through the vehicle's infotainment system, vehicle networking platform, or developer tools to create intelligent scenarios that meet specific usage scenarios.

[0003] In existing technologies, the construction of intelligent scenarios typically relies on OEMs, developers, or professional engineers to conduct user research, define scenarios, configure logic, write scripts, conduct testing and deployment, and perform subsequent maintenance. For example, for scenarios such as "going home from get off work," "driving in the rain," "children riding in the car," and "long-distance driving," triggering conditions, actions, execution order, and control parameters need to be predefined, and then the vehicle system executes them according to fixed logic. While this approach can meet some common needs, its scenario design, configuration, deployment, and optimization processes are highly dependent on manual labor, resulting in low overall efficiency and insufficient scenario flexibility and personalization. Furthermore, the needs raised by users during actual vehicle use are often complex, ambiguous, and dynamically changing. For example, users may express their true needs simply through natural language or behavioral signals such as "I'm a little annoyed driving now," "My child is asleep," or "It's raining today and the traffic is terrible." Traditional implementation methods based on preset scenarios and fixed rules struggle to understand these complex intentions in a timely and accurate manner, and also find it difficult to automatically generate personalized intelligent scenarios that match the current user's needs by combining vehicle status, external environment, user history, and available vehicle capabilities.

[0004] Therefore, it is necessary to provide a new method for the automated design and construction of intelligent scenarios, so that intelligent scenarios can be transformed from engineering activities that rely on manual configuration into intelligent service processes driven by data, large model understanding and automatic planning, thereby reducing the development and maintenance costs of OEMs, improving the efficiency of intelligent scenario construction, and enhancing the vehicle's ability to understand and adapt to complex user needs. Summary of the Invention

[0005] The purpose of this invention is to provide a large model-driven intelligent scene automated design and construction system and method. Through a closed-loop process of vehicle-side data collection, cloud-based big data analysis, large model intent understanding and reasoning planning, automatic generation of scene scripts, vehicle execution and feedback optimization, the intelligent scene is automatically constructed from user needs to executable vehicle control scripts.

[0006] To achieve the above-mentioned technical effects, the present invention adopts the following technical solution: a large model-driven intelligent scene automated design and construction system and method, comprising the following steps:

[0007] A large-scale model-driven intelligent scene automated design and construction system, including vehicle-side components, cloud service platform, and execution module;

[0008] The vehicle terminal includes a data acquisition module and a data upload module. The data acquisition module is used to collect multi-source sensing data, and the data upload module is used to upload the multi-source sensing data to a cloud service platform.

[0009] The cloud service platform includes a big data analytics layer and a big model layer;

[0010] The big data analysis layer is used to process the multi-source sensing data, extract features, identify user behavior patterns, build user profiles, and trigger scene tasks.

[0011] The large model layer is used for intent understanding, scenario reasoning planning, and scenario execution script generation based on scenario tasks, user profiles, vehicle status, environmental conditions, and callable atomic services.

[0012] The execution module is used to receive the scenario execution script and call the atomic service on the vehicle to perform the corresponding action according to the scenario execution script;

[0013] The vehicle terminal is also used to collect user feedback data and execution result data after the scene is executed. The user feedback data and execution result data are used to update the user profile in the big data analysis layer and to optimize the scene generation results of the big model layer.

[0014] Preferably, the multi-source sensing data includes at least one of environmental sensing data, cockpit sensing data, vehicle sensing data, network sensing data, and user historical behavior data.

[0015] Preferably, the environmental perception data includes at least one of the following: outside temperature, humidity, weather, light intensity, road environment, positioning signal, geofence information, and navigation route information;

[0016] The cabin perception data includes at least one of the following: voice commands, gesture commands, touch operations, seat occupancy status, seat pressure, in-vehicle object monitoring signals, in-vehicle temperature, in-vehicle carbon dioxide concentration, and occupant status.

[0017] The vehicle perception data includes at least one of the following: vehicle speed, tire pressure, door status, window status, battery charge, air conditioning status, seat status, media status, and driving mode; the network perception data includes at least one of the following: real-time traffic conditions, cloud-synchronized calendar events, map service data, network connectivity status, and external information service data.

[0018] Preferably, the data upload module is used to format, package, time-series mark, and identify the data source of the multi-source sensing data according to a preset transmission protocol and data specifications, and upload it to the cloud service platform through the vehicle network.

[0019] Preferably, the big data analysis layer includes a data processing module, a calculation and feature module, a user profile and pattern library module, and a scenario task triggering module;

[0020] The data processing module is used to clean, remove anomalies, standardize formats, and align the time sequence of multi-source data for uploaded data.

[0021] The computation and feature module is used to extract real-time state features and statistical features through streaming computation and / or batch processing;

[0022] The user profile and pattern library module is used to generate and update user profile tags and user behavior patterns based on historical and real-time data.

[0023] The scenario task triggering module is used to trigger scenario construction tasks based on real-time status features, user profile tags, and / or user interaction input.

[0024] Preferably, the user profile and pattern library module uses at least one algorithm among cluster analysis, association rule mining, and sequence pattern analysis to identify user behavior patterns and convert the user behavior patterns into structured user profile tags.

[0025] Preferably, the large model layer includes a language and intent understanding module, a scene reasoning and planning module, a scene management module, and a scene optimization module;

[0026] The language and intent understanding module is used to analyze the actual needs expressed by the user through voice, gestures, touch, or action input;

[0027] The scenario reasoning and planning module is used to generate scenario execution scripts based on user profiles, vehicle status, environmental conditions, safety constraints, and callable atomic services.

[0028] The scene management module is used to store, manage versions, enable, disable, update, and distribute scene execution scripts;

[0029] The scenario optimization module is used to optimize the scenario execution script based on user feedback data and execution result data.

[0030] Preferably, the scenario execution script includes at least one of the following: triggering conditions, execution target, execution action, action sequence, atomic service call interface, execution parameters, security constraints, exception handling strategy, and feedback collection rules.

[0031] Preferably, the atomized services include at least one of the following: air conditioning temperature control, seat position adjustment, seat heating or ventilation control, ambient light color switching, window opening or closing, media content playback, navigation route adjustment, and driving mode switching.

[0032] A large-model-driven method for the automated design and construction of intelligent scenarios includes:

[0033] S1: Vehicle-side data collection from multiple sources;

[0034] S2: The data upload module formats, packages, timestamps, and identifies the data source of the multi-source sensing data, and uploads it to the cloud service platform;

[0035] S3: The big data analysis layer of the cloud service platform performs data cleaning, format standardization, time sequence alignment, feature extraction, and state recognition on the multi-source sensing data;

[0036] S4: Identify user behavior patterns based on historical and real-time data, and generate user profile tags and a user profile and pattern library;

[0037] S5: Construct tasks based on scenarios triggered by real-time status characteristics, user profile tags, and / or user interaction input;

[0038] S6: The large model layer generates scenario execution scripts based on scenario construction tasks, user profiles, vehicle status, environmental conditions, and callable atomic services;

[0039] S7: The execution module executes the script based on the scenario and calls the vehicle atomic service to perform the corresponding action;

[0040] S8: The vehicle collects user feedback data and execution result data after the scenario is executed, and updates the user profile and optimizes the large model layer based on the user feedback data and execution result data.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This invention enables the big data analysis layer and the big model to work together, with the big data analysis layer responsible for data processing, feature extraction, pattern mining and user profile construction, and the big model responsible for intent understanding, cross-domain reasoning, scene planning and script generation, thereby forming a well-defined and continuously optimized intelligent scene construction architecture.

[0043] 2. This invention can automatically generate personalized scene execution scripts based on real-time user needs, user profiles, vehicle status, environmental information, and callable atomic services, reducing manual scene design, coding, debugging, and maintenance work, and improving the efficiency of intelligent scene construction.

[0044] 3. This invention forms a closed loop by combining data collected from the vehicle, cloud-based analysis results, large-scale model scene generation results, vehicle execution results, and user feedback data. This enables user profiles and scene patterns to be dynamically updated as user behavior changes, and allows the large-scale model to continuously optimize its intent understanding and scene generation capabilities.

[0045] 4. This invention can transform the traditional highly specialized and engineered intelligent scenario construction process into a user-oriented intelligent service process, reducing the user's usage threshold and the OEM's maintenance costs, and improving the vehicle's ability to adapt to complex and ambiguous user intentions.

[0046] 5. This invention achieves intelligent scene generation and execution through vehicle-cloud collaboration, enabling the cloud to utilize strong computing power for large model inference and scene planning, while the vehicle executes specific atomic services in the script through standardized interfaces, thereby balancing scene generation capabilities and vehicle execution reliability. Attached Figure Description

[0047] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0048] Figure 1 This is a schematic diagram of the vehicle-cloud collaborative architecture of a large model-driven intelligent scene automated design and construction system in one embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the composition of multi-source sensing data in one embodiment of the present invention;

[0050] Figure 3 This is a logical diagram of a large model-driven intelligent scene automated design and construction system in one embodiment of the present invention;

[0051] Figure 4 This is a flowchart illustrating a method for automated design and construction of intelligent scenes driven by a large model, as described in one embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0054] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, all directional indications in this application (such as up, down, left, right, front, back, bottom, etc.) are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indication will also change accordingly. Furthermore, descriptions involving "first," "second," etc., in this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0055] Example 1

[0056] like Figure 1 - Figure 3 As shown, a large-model-driven intelligent scene automated design and construction system includes a vehicle terminal, a cloud service platform, and an execution module.

[0057] The vehicle-side component includes a data acquisition module and a data upload module. The data acquisition module, with user authorization, collects multi-source perception data, including at least one of environmental perception data, cabin perception data, vehicle perception data, network perception data, and user historical behavior data. The data upload module formats, packages, timestamps, and identifies the data source of the multi-source perception data according to a preset transmission protocol and data specifications, and uploads it to a cloud service platform via the vehicle-to-everything (V2X) network.

[0058] The cloud service platform comprises a big data analytics layer and a large model layer. The big data analytics layer performs data cleaning, anomaly removal, format standardization, multi-source data time-series alignment, feature extraction, user behavior pattern recognition, user profile construction, and scenario task triggering on the uploaded data stream. The large model layer performs language and intent understanding, cross-domain reasoning, multi-objective planning, scenario script generation, scenario management, and scenario optimization for scenario tasks.

[0059] The execution module is used to receive scene execution scripts from the vehicle's infotainment system or cloud service platform, and to call atomic services on the vehicle according to the scene execution scripts to perform corresponding actions. The atomic services include at least one of the following: air conditioning temperature control, seat position adjustment, seat heating or ventilation control, ambient lighting color switching, window opening or closing, media content playback, navigation route adjustment, and driving mode switching.

[0060] Furthermore, the big data analytics layer includes a data processing module, a computation and feature module, a user profiling and pattern library module, and a scenario task triggering module. The data processing module is used to clean, remove anomalies, standardize formats, and align multi-source data time series for uploaded data. The computation and feature module is used to calculate key indicators and extract real-time status features through streaming computation and / or batch processing. The user profiling and pattern library module is used to identify user behavior patterns based on historical and real-time data and convert these patterns into structured user profile tags. The scenario task triggering module is used to trigger scenario construction tasks based on real-time status features, user profile tags, and / or user interaction input.

[0061] Furthermore, the large model layer includes a language and intent understanding module, a scene reasoning and planning module, a scene management module, and a scene optimization module. The language and intent understanding module parses the actual needs expressed by the user through voice, gestures, touch, or action input, and generates structured intents. The scene reasoning and planning module performs multi-objective planning by combining user profiles, vehicle status, environmental conditions, safety constraints, and callable atomic services, and generates scene execution scripts. The scene management module stores, manages versions of, enables, disables, updates, and distributes the scene execution scripts. The scene optimization module optimizes the scene execution scripts and the large model generation results based on user feedback data and execution result data.

[0062] Furthermore, the inputs received by the large model layer include a list of callable atomic service capabilities, real-time data, user interaction information, user profile information, and scenario task trigger inputs. The list of callable atomic service capabilities can be in JSON format or other structured formats recognizable by the large model, describing the names, interface identifiers, input parameters, output parameters, and safety constraints of the atomic services currently available for the vehicle to invoke. Real-time data includes vehicle status data and environmental data. User interaction information includes user voice phrases, text phrases, gesture input, or touch input. User profile information includes user comfort preferences, safety preferences, and behavioral habit tags.

[0063] The large model layer first performs semantic parsing on user interaction information to identify the primary intent expressed by the user. Then, it combines user profiles to identify secondary or implicit intents, and extracts conditional intents based on vehicle status and environmental data, thus outputting structured user intents. For example, when the ambient temperature is 35℃, and the user profile shows that the user has a habit of using automatic air conditioning and prefers a temperature of 18℃, the large model layer can interpret the user's expression "it's a little hot" as an intent to lower the interior temperature. When a child passenger is also detected in the vehicle, the large model layer can adjust the target air conditioning temperature to a more suitable temperature for the child, taking into account the child's comfort and safety needs.

[0064] When user interaction information contains multiple intents, the large model layer assigns weights to these intents. The intent weight corresponding to the user's active command is greater than the implicit intent weight inferred from the user profile. For primary and implicit intents that can be satisfied simultaneously, the large model layer establishes an association relationship. For conflicting and mutually exclusive intents, the large model layer establishes an association relationship and determines the priority intent target to be executed based on the weight results. Through the above processing, the large model layer can transform complex, ambiguous, or multi-objective user expressions into scene task trigger inputs that can be used for script generation.

[0065] Furthermore, the scenario execution script includes at least one of the following: triggering conditions, execution target, execution action, action sequence, atomic service call interface, execution parameters, security constraints, exception handling strategy, and feedback collection rules.

[0066] Furthermore, the scenario execution script also includes a trigger condition section, a status condition section, and an execution instruction section. The trigger condition section describes the script startup conditions, the status condition section describes the security check conditions before or during script execution, and the execution instruction section describes the vehicle atomic service to be invoked and its execution parameters.

[0067] Specifically, the triggering condition section includes a number of triggering conditions (triggerNum) and a set of triggering conditions (triggers). The set of triggering conditions (triggers) includes at least one triggering condition node (trigger0 to triggerL-1). Each triggering condition node includes an interface name (interfaceId), a triggering condition comparison operator (triggerComparisonOperator), and a triggering condition parameter (triggerParameter). The interface name (interfaceId) corresponds to the vehicle atomicity service. The triggering condition comparison operator includes at least one of <, ≤, >, ≥, ≠, and =. The triggering condition parameter is used to compare the triggering condition with the vehicle status value or environment status value returned by the corresponding interface.

[0068] The status conditions section includes the number of check conditions, conditionNum, and the condition check set, conditions. The condition check set, conditions, includes at least one condition node, condition0 to conditionN-1. Each condition node includes an interface name (interfaceId), a condition comparison operator (comparisonOperator), and a condition parameter (conditionParameter). The condition nodes are used to determine whether the vehicle status, environment status, user status, or security status meets the script execution requirements.

[0069] The execution instruction section includes the number of execution instructions (executionNum) and the set of execution instructions (executions). The execution instruction set (executions) includes at least one execution instruction node (execution0 to executionM-1), and each execution instruction node includes an interface name (interfaceId) and an execution parameter (executionParameter). The execution module calls the corresponding vehicle atomic service based on the interface name (interfaceId) and uses the execution parameter (executionParameter) as the execution target of that atomic service. In this implementation, the key parameters of the script format are shown in the table below.

[0070]

[0071] Furthermore, the user feedback data includes at least one of the following: user confirmation execution scenario, user refusal execution scenario, user active interruption scenario, user manual adjustment of scenario parameters, repeated triggering of similar requirements, and changes in behavior after execution; the execution result data includes at least one of the following: scenario execution success or failure, vehicle status change, execution time, interface call result, and exception information.

[0072] like Figure 4As shown, the present invention also provides a method for automated design and construction of intelligent scenes driven by a large model, comprising the following steps:

[0073] S1: The vehicle collects multi-source sensing data through the data acquisition module;

[0074] S2: The data upload module formats, packages, timestamps, and identifies the data source of the multi-source sensing data, and uploads it to the cloud service platform;

[0075] S3: The big data analysis layer of the cloud service platform performs data cleaning, format standardization, time sequence alignment, feature extraction, and state recognition on the multi-source sensing data;

[0076] S4: Based on historical and real-time data, identify user behavior patterns and generate user profile tags and a user profile and pattern library;

[0077] S5: Construct tasks based on scenarios triggered by real-time status characteristics, user profile tags, and / or user interaction input;

[0078] S6: The large model layer constructs tasks based on the scenario, and combines user profiles, vehicle status, environmental conditions, safety constraints and callable atomic services to perform intent understanding, cross-domain reasoning and multi-objective planning to generate scenario execution scripts;

[0079] S7: The execution module executes the script based on the scenario and calls the vehicle atomic service to perform the corresponding action;

[0080] S8: The vehicle collects user feedback data and execution result data after the scenario is executed, and updates the user profile and optimizes the large model layer based on the user feedback data and execution result data.

[0081] Example 2

[0082] Using the system described in Example 1, this example provides a method for automated design and construction of intelligent scenes driven by a large model. The method includes the following steps:

[0083] The first step involves the vehicle-side data acquisition module continuously collecting multi-source sensor data, with user authorization. This multi-source sensor data includes environmental perception data, cabin perception data, vehicle perception data, network perception data, and user historical behavior data. Specifically, environmental perception data includes outside temperature, weather, road conditions, location signals, and navigation route information; cabin perception data includes voice commands, gesture commands, touch operations, seat occupancy status, seat pressure, in-vehicle object monitoring signals, and in-vehicle carbon dioxide concentration; vehicle perception data includes vehicle speed, tire pressure, door status, window status, battery level, air conditioning status, media status, and driving mode; and network perception data includes real-time traffic conditions, cloud-synchronized calendar events, and map service data. The data acquisition module performs preliminary cleaning, filtering, and formatting of the raw data and then sends the processed data to the data upload module.

[0084] The second step involves the data upload module receiving data packets from the data acquisition module, repackaging the data packets according to the preset transmission protocol and data specifications, adding time stamps and data source identifiers, and then uploading them to the cloud service platform via the vehicle network to form a unified and orderly raw data stream.

[0085] The third step involves the cloud service platform's big data analytics layer performing in-depth processing on the raw data stream. Specifically, the big data analytics layer first cleans the raw data stream, removes anomalies, standardizes the format, and aligns the time sequence of multi-source data. Subsequently, it calculates key indicators and extracts real-time status features through a combination of streaming computing and batch processing. These real-time status features include at least one of the following: the vehicle is in a congested area, the carbon dioxide concentration inside the vehicle is high, the user is continuously adjusting the air conditioning temperature, and the user is playing a fixed type of music during a specific time period.

[0086] The fourth step involves the big data analytics layer identifying user behavior patterns based on historical and real-time data using at least one algorithm among clustering analysis, association rule mining, and sequence pattern analysis. These patterns are then transformed into structured user profile tags and stored in the user profile and pattern library module. The user profile tags include at least one of the following: commuting preferences, temperature preferences, music preferences, congestion-related emotional tendencies, fatigue driving risk, children's riding habits, and in-vehicle air quality sensitivity. The user profile and pattern library module supports dynamic updates, allowing user profiles to evolve as user behavior changes.

[0087] The fifth step involves continuously inputting multi-source real-time data streams into the big data analysis layer. The big data analysis layer matches the real-time status characteristics with the user profile tags to a pattern library, triggering a scenario building task. Alternatively, after the vehicle system receives interactive information input by the user through voice, gestures, touch, or actions, it triggers the language and intent understanding module of the big model layer. The language and intent understanding module then parses the user's actual needs and triggers a scenario building task.

[0088] Step six involves the large model layer performing language understanding, intent recognition, cross-domain reasoning, and multi-objective planning for the scene construction task. Specifically, the large model layer comprehensively considers user preferences, vehicle status, environmental conditions, safety constraints, and callable atomic services to plan a scene execution script that matches the user's actual needs. The scene execution script includes at least one of the following: triggering conditions, execution objectives, execution actions, action sequence, atomic service call interfaces, execution parameters, safety constraints, exception handling strategies, and feedback collection rules.

[0089] When user interaction information is broken down into multiple intents, the large model layer assigns weights to these intents, with the intent corresponding to the user's active command having a higher weight than the intent inferred from the user profile. For primary intents and implicit intents that can be satisfied simultaneously, an association is established. For mutually exclusive intents that conflict with each other, an association is established, and the priority intent target is determined based on the weight results.

[0090] The large model layer further plans scenario execution scripts that match actual user needs based on structured user intent, user profiles, vehicle status, environmental conditions, safety constraints, and a list of callable atomic service capabilities. These scenario execution scripts include trigger conditions, state conditions, and execution actions; where trigger conditions and state conditions correspond to vehicle status, environmental status, or atomic service interface parameters, and execution actions correspond to vehicle atomic services such as air conditioning temperature adjustment, music playback, seat adjustment, window control, ambient lighting control, or navigation adjustment.

[0091] After the large model layer generates the scene execution script, it performs self-verification on the scene execution script. This self-verification includes verifying the completeness of necessary script parameters and logical meaning. Figure 1 Consistency checks; among them, the script necessary parameter completeness check is used to determine whether script elements are missing, whether fields are standardized, whether the format is consistent, and the logical meaning. Figure 1 Consistency checks are used to determine whether the script matches the user's intent and the input triggered by the scenario task.

[0092] Step 7: The cloud service platform distributes the scenario execution script to the vehicle-side execution module via a secure interface. The execution module then calls the atomic services on the vehicle according to the scenario execution script to control the air conditioning, seats, windows, ambient lighting, media, navigation, or driving mode to perform corresponding actions.

[0093] When determining the triggering and state conditions in the script, the execution module calls the corresponding atomic service to obtain the vehicle state value, environment state value, or user state value, and determines whether the script triggering and state conditions are met based on the corresponding comparison operator and parameters. If the triggering and state conditions are met, the execution module continues to parse and execute the corresponding action, and calls the corresponding vehicle atomic service to execute the action based on the interfaceId and executionParameter in the execution instruction.

[0094] Furthermore, when executing the scenario execution script, the execution module also performs execution verification based on the pre-set safety constraints of the callable atomic services on the vehicle side. For example, when the vehicle is not in P gear, the execution action corresponding to the electric front trunk opening service does not meet the safety constraints. The execution module skips or ignores this execution action, or returns a prompt message to the user indicating that the execution conditions are not met. Through the above vehicle-side execution verification, it is possible to prevent some actions in the scenario execution script from being executed under unsafe conditions.

[0095] Step 8: After the scenario execution is complete, the data acquisition module collects user feedback data and execution result data. The user feedback data includes user-initiated interruption, manual adjustment of scenario parameters, confirmation of execution, refusal of execution, or subsequent behavior adjustments. The execution result data includes whether the scenario execution was successful or failed, execution time, vehicle status changes, and API call results. The big data analytics layer updates the user profile and pattern library based on the feedback data, and the big model layer optimizes intent understanding and scenario generation results based on the feedback data, making subsequent scenario construction more accurate.

[0096] Through the above settings, this embodiment not only illustrates the input from which the large model generates scene execution scripts, but also further defines the semantic parsing, intent recognition, cross-domain reasoning, multi-objective planning, script format generation, and script verification processes of the large model. Since the triggering conditions, state conditions, and execution actions in the scene execution script all correspond to the vehicle's atomic service through interface names, the vehicle-side execution module can complete condition judgments and service calls based on script node parameters. This allows the large model's generated results to be transformed into vehicle-side executable control scripts, avoiding a focus solely on abstract intent understanding or natural language generation, thus improving the feasibility and verifiability of the technical solution.

[0097] Through the above steps, this embodiment forms a closed-loop intelligent scene construction process of "data acquisition - data upload - feature extraction - profile construction - task triggering - large model inference - scene deployment - scene execution - feedback optimization". It can transform the traditional intelligent scene construction method that relies on manual design, configuration and maintenance into an intelligent service process that is driven by data and automatically completed by large models.

[0098] Example 3

[0099] This embodiment uses a scenario of emotional relief in a congested road section as an example to illustrate the specific application process of the present invention.

[0100] During vehicle operation, the data acquisition module continuously collects data on vehicle speed, navigation route, real-time traffic conditions, in-vehicle temperature, media playback status, user history, and user interaction. The data upload module then uploads this data to the cloud service platform after adding time-series tags and data source identifiers.

[0101] After processing the uploaded data, the big data analytics layer of the cloud service platform identifies that the current time is during the evening rush hour, the road is severely congested, the average vehicle speed is below a preset threshold, and the user profile tags show that the user is prone to irritability during congestion periods and has repeatedly lowered the air conditioning temperature and played soothing music in their historical behavior. Based on these characteristics and user profile tags, the scenario task triggering module initiates a task to construct a congestion-related mood-relieving scenario.

[0102] The language and intent understanding module and the scene reasoning and planning module of the large model layer combine user profiles, vehicle status, environmental conditions, and callable atomic services to determine the scene objectives as alleviating congestion anxiety, maintaining cabin comfort, and reducing driver fatigue. Subsequently, the large model layer generates a scene execution script, which includes: adjusting the air conditioning temperature to the user's preferred temperature, adjusting the fan speed to a comfortable level, switching the ambient lighting to a soft color, playing a playlist of commonly used soothing music, providing navigation detour suggestions, and activating the seat massage function when the vehicle is in a safe state.

[0103] The scenario execution script is sent to the vehicle execution module via a secure interface. The vehicle execution module then calls atomic services such as air conditioning, ambient lighting, media playback, navigation, and seat massage according to the script, and displays the scenario execution status to the user through the vehicle's infotainment system.

[0104] If the user turns off the music, cancels the seat massage, or readjusts the air conditioning temperature during the process, the data acquisition module collects this feedback and uploads it to the cloud service platform. Based on this, the big data analytics layer updates the user profile and pattern library, and the big model layer adjusts the execution parameters when generating similar scenarios in the future, making subsequent congestion relief scenarios more consistent with the user's actual preferences.

[0105] The specific embodiments of the present invention have been described above. Based on the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention.

Claims

1. A large-model-driven intelligent scene automated design and construction system, characterized in that, This includes the vehicle-side components, cloud service platform, and execution modules; The vehicle terminal includes a data acquisition module and a data upload module. The data acquisition module is used to collect multi-source sensing data, and the data upload module is used to upload the multi-source sensing data to a cloud service platform. The cloud service platform includes a big data analytics layer and a big model layer; The big data analysis layer is used to process the multi-source sensing data, extract features, identify user behavior patterns, build user profiles, and trigger scene tasks. The large model layer is used for intent understanding, scenario reasoning planning, and scenario execution script generation based on scenario tasks, user profiles, vehicle status, environmental conditions, and callable atomic services. The execution module is used to receive the scenario execution script and call the atomic service on the vehicle to perform the corresponding action according to the scenario execution script; The vehicle terminal is also used to collect user feedback data and execution result data after the scene is executed. The user feedback data and execution result data are used to update the user profile in the big data analysis layer and to optimize the scene generation results of the big model layer.

2. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The multi-source sensing data includes at least one of environmental sensing data, cockpit sensing data, vehicle sensing data, network sensing data, and user historical behavior data.

3. The large-model-driven intelligent scene automated design and construction system according to claim 2, characterized in that, The environmental perception data includes at least one of the following: outside temperature, humidity, weather, light intensity, road environment, positioning signal, geofence information, and navigation route information; The cabin perception data includes at least one of the following: voice commands, gesture commands, touch operations, seat occupancy status, seat pressure, in-vehicle object monitoring signals, in-vehicle temperature, in-vehicle carbon dioxide concentration, and occupant status. The vehicle perception data includes at least one of the following: vehicle speed, tire pressure, door status, window status, battery charge, air conditioning status, seat status, media status, and driving mode. The network-sensing data includes at least one of the following: real-time traffic conditions, cloud-synchronized calendar events, map service data, network status, and external information service data.

4. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The data upload module is used to format, package, time-series mark, and identify the data source of the multi-source sensing data according to a preset transmission protocol and data specifications, and upload it to the cloud service platform through the vehicle network.

5. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The big data analysis layer includes a data processing module, a calculation and feature module, a user profile and pattern library module, and a scenario task triggering module; The data processing module is used to clean, remove anomalies, standardize formats, and align the time sequence of multi-source data for uploaded data. The computation and feature module is used to extract real-time state features and statistical features through streaming computation and / or batch processing; The user profile and pattern library module is used to generate and update user profile tags and user behavior patterns based on historical and real-time data. The scenario task triggering module is used to trigger scenario construction tasks based on real-time status features, user profile tags, and / or user interaction input.

6. The large-model-driven intelligent scene automated design and construction system according to claim 5, characterized in that, The user profile and pattern library module uses at least one algorithm among cluster analysis, association rule mining, and sequence pattern analysis to identify user behavior patterns and convert the user behavior patterns into structured user profile tags.

7. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The large model layer includes a language and intent understanding module, a scene reasoning and planning module, a scene management module, and a scene optimization module; The language and intent understanding module is used to analyze the actual needs expressed by the user through voice, gestures, touch, or action input; The scenario reasoning and planning module is used to generate scenario execution scripts based on user profiles, vehicle status, environmental conditions, safety constraints, and callable atomic services. The scene management module is used to store, manage versions, enable, disable, update, and distribute scene execution scripts; The scenario optimization module is used to optimize the scenario execution script based on user feedback data and execution result data.

8. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The scenario execution script includes at least one of the following: triggering conditions, execution target, execution action, action sequence, atomic service call interface, execution parameters, security constraints, exception handling strategy, and feedback collection rules.

9. The large-model-driven intelligent scene automated design and construction system according to claim 1, characterized in that, The atomized services include at least one of the following: air conditioning temperature control, seat position adjustment, seat heating or ventilation control, ambient lighting color switching, window opening or closing, media content playback, navigation route adjustment, and driving mode switching.

10. A method for automated design and construction of intelligent scenes driven by a large model, characterized in that, The method, applied to the large-model-driven intelligent scene automated design and construction system according to any one of claims 1 to 9, comprises: S1: Vehicle-side data collection from multiple sources; S2: The data upload module formats, packages, timestamps, and identifies the data source of the multi-source sensing data, and uploads it to the cloud service platform; S3: The big data analysis layer of the cloud service platform performs data cleaning, format standardization, time sequence alignment, feature extraction, and state recognition on the multi-source sensing data; S4: Identify user behavior patterns based on historical and real-time data, and generate user profile tags and a user profile and pattern library; S5: Construct tasks based on scenarios triggered by real-time status characteristics, user profile tags, and / or user interaction input; S6: The large model layer generates scenario execution scripts based on scenario construction tasks, user profiles, vehicle status, environmental conditions, and callable atomic services; S7: The execution module executes the script based on the scenario and calls the vehicle atomic service to perform the corresponding action; S8: The vehicle collects user feedback data and execution result data after the scenario is executed, and updates the user profile and optimizes the large model layer based on the user feedback data and execution result data.

Images